Variables related to vegetation species populations are considered essential for assessing how tundra ecosystems will respond to global change, yet monitoring approaches using available optical satellite imagery do not fully capture plant biodiversity change in the Arctic. High spectral resolution data are a potential solution to the challenges of differentiating spectrally similar tundra vegetation communities as well as estimating biophysical variables such as biomass, carbon fluxes, and green percent cover.

We explore the use of visible and near-infrared (VNIR) spectroscopy for the estimation of plot-scale plant biodiversity in three circumpolar Arctic sites, Toolik Lake, Alaska, Herschel Island, Yukon Territory, and the Yamal Peninsula, Siberia. Spectral measurements were collected near peak season in a combination of monitoring plots and transects depending on the site. Spectral signatures were compared to biodiversity data extracted from point-framing or Braun-Blanquet relevé estimates collected in the same locations. We will use these data to test the relationship between spectral reflectance and biodiversity data at various spectroradiometer fields of view (FOV) and biodiversity estimates.

These analyses will allow us to identify how and if spectral signatures can be used as a proxy for plant community biodiversity in addition to testing how spatial scale influences the applicability of VNIR spectroscopy as a monitoring tool for tundra plant biodiversity in the Arctic.

Area A - Board 29

Ecosystem functions in grasslands are essential for the survival of plant and animal species and vital in sustaining human life. A number of studies have shown that ecosystem functions are strongly linked to locally measured plant traits and their variations. This relationship between trait diversity, called plant functional diversity, and ecosystem functions is likely to be scale dependent and varying within the phenological cycle. Further, trait diversity estimation on large spatial scales might add crucial information related to resilience of an ecosystem. A faster recovery of disturbed area depending on the availability of surrounding ecosystem functions has to be expected. However, capturing the trait variations at distinct spatial scales and temporal stages and in differently managed grassland remains a cost- and time-efficiency nightmare. Modern remote sensing systems bear the potential to close the gap produced by a limited availability of trait measurements. Optical sensors benefit from the unique light reflectance properties of plant traits. In particular, Sentinel-2 datasets provide a complete landscape picture with high revisit time. Based on Sentinel-2 data, we quantified plant traits in grasslands of the Swiss National Park and the agricultural landscape in its surroundings. For trait quantification we took advantage of radiative transfer modelling and vegetation indices. Different aspects of functional diversity were quantified in space and time. Applied to our study area, it was possible to distinguish the impact of different fertilization and grazing intensities on plant traits and functional diversity on a regional scale and in time. Further, we demonstrated the importance of quantifying large-scale functional diversity from space due to discrepancy with a locally measured diversity. Functional diversity on a local scale is affected by nutrient availability, grazing and mowing intensities. On a regional scale, control parcel size and the uniformity of agricultural practices and policies over large areas diversity. Variance in time of functional diversity is characterized by mowing events. Overall, plant trait diversity retrieved from Sentinel-2 datasets produces evidence for the evaluation of protective and agricultural measurements on extended spatial scales, which are important for conservation and restoration decision-making. Further, due to high revisit time of Sentinel-2, the method is suitable for incorporation into monitoring schemes.

Area A - Board 30

The aim of this research is to assess the feasibility of retrieving plant functional traits from Sentinel-3 data by using combined vegetation-atmosphere inversions of radiative transfer models.
Sentinel-3 is a constellation of ESA’s satellites (Sentinel-3A and 3B) that have roughly 30 bands on two sensors (Ocean and Land Colour Instrument, OLCI and Sea and Land Surface Temperature Radiometer, SLSTR) in the visible, near- and mid-infrared regions. The revisit time of about one day makes Sentinel-3 suitable for producing time series and monitoring the ecosystem state. However, the measurements provided by Sentinel-3 are strongly affected by the atmosphere. Atmospheric correction is usually carried out before surface parameter retrievals, but we attempted to retrieve properties from OLCI and SLSTR from the radiative transfer models SCOPE (Soil Canopy Observation, Photochemistry and Energy fluxes) and 6S [1,2] so the vegetation and atmospheric parameters are simultaneously modelled. SCOPE is a radiative transfer and micro-meteorological model that simulates reflectance spectra at leaf and canopy levels as well as photosynthesis and the components of the energy balance (net radiation, sensible and latent heat flux) for soil an canopy [3]. We developed an algorithm based on the coupling of the optical module of SCOPE with the atmospheric radiative transfer model 6S to retrieve plant traits (chlorophyll content, leaf area index, water content) from Sentinel-3 measurements. Clouded images, based on a cloud band of SLSTR product, and images without vegetation, based on a low value of the normalized difference vegetation index (NDVI), were excluded from the analysis. The retrieved plant traits were further used as an input for the complete SCOPE model to produce ecosystem fluxes.
Ground measurements from a number of European FLUXNET eddy-covariance towers were used for the validation of the simulated ecosystem fluxes. The plant functional types represented by those towers included savannahs, croplands, grasslands, evergreen broadleaf and needleleaf forests and deciduous forests. To reduce the noise in the data, we compared 3-days mean fluxes.
The project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 721995.
References
1. Vermote, E. F.; Tanre, D.; Deuze, J. L.; Herman, M.; Morcette, J.-J. Second Simulation of the Satellite Signal in the Solar Spectrum, 6S: an overview. IEEE Trans. Geosci. Remote Sens. 1997, 35, 675–686, doi:10.1109/36.581987.
2. Wilson, R. T. Py6S: A Python interface to the 6S radiative transfer model. Comput. Geosci. 2013, 51, 166–171, doi:10.1016/j.cageo.2012.08.002.
3. Van der Tol, C.; Verhoef, W.; Timmermans, J.; Verhoef, A.; Su, Z. An integrated model of soil-canopy spectral radiances, photosynthesis, fluorescence, temperature and energy balance. Biogeosciences 2009, 6, 3109–3129, doi:10.5194/bg-6-3109-2009.

Area A - Board 31

The Chilean-Peruvian coastal desert is one of the most extreme geo-ecosystems worldwide. In its core zone, the average annual precipitation in the form of episodic rainfall events falls below 2 mm. Hence, plant growth is limited to so-called fog oases, where ecologically specialized plants, like Tillandsia spp., cover their entire water demand by combing out water from the coastal fog that regularly occurs from the Pacific Ocean during night. In consequence of their strong dependency on the coastal fog, Tillandsia spp. are highly vulnerable to changes in its frequency and intensity and may therefore serve as a bioindicator for climate change. Yet, the main dynamics of these fog geo-ecosystems, locally known as Lomas or Loma-formations, are still poorly understood on both contextual and spatial scale.
Against this background, a multi-scale remote sensing approach for the detection, analysis and monitoring of the distribution patterns of the fog geo-ecosystems in the Chilean-Peruvian coastal desert was developed. On a regional level, WorldView-3 data were used to develop semi-automated algorithms to detect the spatial extent of Tillandsia-Lomas in the entire study area. To obtain a quantitative description of the spatial structural dimensions, the local distribution characteristics of selected ecosystem sites in northern Chile were then examined in more detail. Data from unmanned aircraft systems were used to create orthophotos and digital surface models with a spatial resolution in the sub-decimeter range. From these products, several structural characteristics (e.g. coverage ratio, cover patterns) of the plant stands were derived, which allow a systematic and accurate spatial description. Additionally, in cooperation with the Centre for Organismal Studies (COS) Heidelberg, the endemic Tillandsia landbeckii was examined regarding vitality and genetic diversity. The relationship between the distribution of the Tillandsia-Lomas and the coastal fog (occurrence, frequency, and intensity) were analyzed via a correlation with remotely sensed stratocumulus cloud detections and on site fog water yields, which were measured by standard fog collectors and additional climatic sensors. Finally, the results of the local inventory analyses were extrapolated to the regional level via WorldView-3 data and topographical models.
The presented method is an integrative approach of several remote sensing products with multi-level hierarchy, complemented by in-situ observations. The results contribute to a better understanding of the interaction between atmosphere and biosphere in the Chilean-Peruvian coastal desert and may serve as the basis for protecting the fog geo-ecosystems and their endemic species. Moreover, the method allows for a systematic detection and monitoring of future changes in the distribution patterns of the Tillandsia-Lomas as an indicator for climate change as well as for the estimation of using fog water as local sustainable water resource.

Area A - Board 32

Drylands constitute one of the main ecosystems worldwide, being estimated to cover half of the global land surface by the end of this century. Considered one of the most vulnerable areas to the global change, drylands are composed by sparse vegetation and soil interspaces covered by biocrusts, a combination of communities including mosses, lichens, liverworts, algae and cyanobacteria. Being fundamental elements in drylands, these communities have been suggested to be used asindicators of ecosystem functioning in soils, which is why their loss of biodiversity is getting special attention recently. Albeit measuring biodiversity as an indicator of ecosystem health has been recognized worldwide by major initiatives (e.g. GEO BON, WCRP), estimating it from field data presents drawbacks, mainly related with time and costs. This underlies the need to develop new techniques for this.
In this contribution, we examined the potential of multispectral drone imagery to capture biodiversity in biocrusted surfaces of a drylands system in Aranjuez, central Spain. During a field campaign, different compositions of biocrusts (i.e. lichens and mosses) and bare soil were sampled, plots of interest were characterized, and multispectral images (Parrot Sequoia, senseFly, Switzerland) were acquired at the same time. Optical properties of the biocrust samples were measured in laboratory by means of a hyperspectral imaging spectrometer (Hyperspec VNIR, Headwall Photonics, USA) covering the VIS-NIR spectrum (400-1000 nm). The fractional cover was then calculated using an object-based image analysis (OBIA) approach.
At first, working at ground level, and based on their fractional cover, we calculated species richness, evenness (using Pielou’s index) and diversity (using the exponential Shannon index, an entropy measure that combines species richness and evenness) as biocrust biodiversity indicators. We then calculated the spectral diversity as the coefficient of variation of the mean spectra taken with the hyperspectral imaging spectrometer, and we related it with the biodiversity of the samples.
Based on the results obtained at ground level we then tested the possibility to use the spectral diversity calculated from the drone multispectral imagery to characterise the spatial patterns of biocrust diversity at landscape level. To focus only on biocrusted surfaces, we masked the vegetation from the images, using an OBIA approach, and the bare soil areas, based on their spectral properties. The spectral diversity was then calculated on the biocrusted surfaces identified on the drone images using different kernel sizes and compared to the results obtained at ground level.
Results obtained at ground level demonstrated the existence of a significant relation between spectral heterogeneity and biodiversity. Upscaling the same methodology at landscape level showed the potential of using drone multispectral imagery to monitor biodiversity patterns at a wider scale.

Area A - Board 33

Accurate land surface retrieval is climbing towards a golden-age with the launch of several new higher spatial and temporal resolution satellites such the Sentinel series. With the increase of Earth Observation (EO) data of this calibre, the implementation of data assimilation methods for retrieving relevant land surface parameters can become more robust and feasible, especially with the use of a priori data.
Currently, robust and accurate land surface retrieval has not been fully satisfied, because most estimation schemes apply 1) a single-sensor single-scale approach, and 2) require specific key-variables to be ‘guessed’. This is due to the large amount of observational information required to accurately retrieve relevant parameters. Consequently, many schemes assume specific variables to be constant or not present; subsequently leading to more uncertainty.
In this aspect, the MULTIscale SENTINEL land surface information retrieval Platform (MULTIPLY) platform was created. MULTIPLY couples a variety of RS sources with Radiative Transfer Models (RTM) over varying spectral ranges using data-assimilation to estimate geophysical parameters. In addition, MULTIPLY also uses prior information about the land surface to constrain the retrieval problem, including functional plant traits.
This research aims to improve vegetation parameter retrieval using MULTIPLY with plant trait priors in three stages: 1) Meta-analysis on existing plant-trait data(bases) to derive trait-trait covariance and variability boundaries for model constraints, 2) A collection of in-situ plant traits in various ecosystems over growing seasons to model trait-trait covariance during phenological changes, 3) Compare performance of these priors with multi-spectral and hyperspectral data. This section focuses on the first stage priors, testing retrievals with these stage 1 priors and an overview of future stages current progress. Plant traits of interest include those which overlap between those in RTM models, and those which can also be related to individual performance or fitness of species of plants (i.e. chlorophyll, leaf water content, leaf mass per area, leaf dry matter content, etc.)
For stage 1, plant traits were collected for a meta-analysis from databases such as TRY, Ecosis, and individual collaborators. Plant traits such as chlorophyll, carotenoids, anthocyanins, phenols, leaf water, leaf dry matter content, leaf mass per area (LMA) were considered. Principle component analysis (PCA) and a covariance matrix were generated for use as priors. Then MULTIPLY was used to retrieve several plant traits in a forest validation site in the Netherlands (Speulderbos).

Area A - Board 34

It is well-known that the temporal dimension of Earth observation programmes, such as Landsat or Copernicus, allows new insights into temporal dynamics of the Earth’s surface. While a transformation of pixel values into land-cover or land-use classes is always difficult and error-prone, some applications benefit from shallower but less error-prone and more consistent information about the variability or stability of land cover. Examples include variation in vegetation cover or surface water frequency. The LIVID (Long Image Time-Series Variability Indicator Description) approach applies a fully automated workflow to calculate the overall variability of semantic content for uniform objects of any size between multiple images. This type of temporal analysis produces a single map of index values based on user-defined weights of their category changes over time, thus indicating regions with high relative variability or stability. In the applied example presented here, the calculated index is based on semantically enriched, data-derived categories and user-defined weights pertaining to natural resource depletion.
The workflow consists of the following steps:
1. Calculation of information layers for each image (time stamp)
2. Creating objects (here: regular objects by image tessellation) as information units
3. Calculating a weighted index for each object based on each information layer
4. Calculating the overall temporal variability
We used the SIAM™ software to pre-process each Sentinel-2 multi-spectral image into pre-classified information layers in an automated way (i.e. without any user-defined parameters) (Baraldi et al. 2018a, 2018b). These information layers are used as input for the rest of the workflow using an object-oriented data modelling concept as implemented in the eCognition software (Trimble Geospatial). The legend consists of 33 categories and is purely mapped based on each pixel’s colour information (i.e. spectral response). Example colour categories are “green as strong vegetation” or “dark blue as deep water or shadow”. However, it needs to be considered throughout the whole analysis that colour, i.e., spectral reflectance, might be associated to more than one land use or land cover class and vice versa. Alternative information layers or scene classification maps might be used as well, if so desired.
A user-defined tessellation of the image footprint, e.g. using uniform hexagons or squares, is generated to create objects as primary analysis unit. The size and shape of the objects remains constant through the temporal dimension of the analysis. The categories are weighted based on user requirements to calculate an index for each object at each time-stamp. The index values are then collectively used to describe the variability over time.
In our example, we calculated the weighted natural depletion index by Hagenlocher, Lang, and Tiede (2012) by aggregating selected categories of the 33 categories, which are associated with natural resources and weighing the aggregated groups based on the same expert recommendations. For example, categories indicating strong vegetation are weighted higher than water or bare soil, since vegetation is valued higher in this particular natural depletion index. For each time-stamp, each object receives an index value ranging between 0 and 1 based on the relative area of each category within each object, multiplied by the respective user-defined weights.
We then calculate the Cartesian distances between each object’s temporally successive index values, where the time is on the y-axis and the index values are on the x-axis. The sum of these distances is then normalised using the Cartesian distance between the first and last time-stamp. This measure of relative variability was introduced by Braun and Hochschild (2017) for SAR images, but has not yet been used for automatically pre-classified optical images.
In contrast to other approaches, this approach leverages a fully automated process for pre-classification using SIAM™ and variability calculation within eCognition, but limits the result to information on each object’s variability through time. Nevertheless, for some applications this is a valuable augmentation of the toolboxes. For example, it might be used as pre-analysis to identify areas that are worth investigating further with additional methods, e.g. dynamic time-warping, or selecting ground reference points. In data cubes, this can be used to index the temporal domain, which might be used to optimise tiling schemes and queries, similar to relational databases. It might also provide additional evidence in combination with other tools and interpretation of temporal changes.

References
Baraldi, Andrea, Michael L. Humber, Dirk Tiede, and Stefan Lang. 2018a. “GEO-CEOS stage 4 validation of the Satellite Image Automatic Mapper lightweight computer program for ESA Earth observation level 2 product generation - Part 1: Theory.” Cogent geoscience 4 (1): 1–46. doi:10.1080/23312041.2018.1467357.
Baraldi, Andrea, Michael L. Humber, Dirk Tiede, and Stefan Lang. 2018b. “GEO-CEOS stage 4 validation of the Satellite Image Automatic Mapper lightweight computer program for ESA Earth observation level 2 product generation - Part 2: Validation.” Cogent geoscience 4 (1): 1467254. doi:10.1080/23312041.2018.1467254.
Braun, Andreas, and Volker Hochschild. 2017. “A SAR-Based Index for Landscape Changes in African Savannas.” Remote Sensing 9 (4): 359. doi:10.3390/rs9040359.
Hagenlocher, Michael, Stefan Lang, and Dirk Tiede. 2012. “Integrated assessment of the environmental impact of an IDP camp in Sudan based on very high resolution multi-temporal satellite imagery.” Remote Sensing of Environment 126: 27–38. doi:10.1016/j.rse.2012.08.010.

Area A - Board 35

The scope of our research is Emiliania huxleyi – the most abundant and productive calcifying phytoplankton in the World Ocean. In previous studies we implemented quantitative evaluation of its impact on global oceanic carbon cycle for polar and subpolar latitudes, where Arctic amplification is distinctively present, using remote sensing data from 1998 to 2016 (modern high-end satellite measurements era).
Our goal is distinguishing and prioritizing factors that affect forming of E. huxleyi blooms and their areas (S), which sometimes cover hundreds of thousands or even millions square kilometers. For the aforesaid latitudes we analyzed dynamics of S depending on values of certain physical factors such as sea surface temperature and salinity (SST & SSS), photosynthetic available radiation (PAR), water current speed (WCS) and Ekman layer depth (ELD) [derivable from above water surface wind speed (WS)].
Our research based on satellite data is pioneering: unlike numerous laboratory and shipborne/ in situ studies of the influence of individual environmental parameters on E. huxleyi blooms, the present study reveals a joint impact of multiple physical factors within huge spatial and temporal coverage (1998-2016).
We present results of statistical analysis of aforementioned five parameters, i.e. SST, SSS, PAR, WCS, and ELD for six target seas (North, Norwegian, Labrador, Greenland, Barents and Bering seas). The analysis includes descriptive statistics of blooming regions and results of implementing a machine learning algorithm Random Forest Classifier, based on a Decision tree method. Latter was used for prioritizing bloom-forming factors according to their impact on dynamic of E. huxleyi blooming.
The study shows that the optimal ranges of affecting parameters vary for each targeted sea, and the values of parameters differ between inside and beyond the bloom area within the sea. It brings us to the conclusion that a certain complex of values of factors is necessary for this alga to start blooming. As a result of modeling, a particular affecting factor or a combination of them were established for each sea. Results of this study give us a) a better understanding of the causes of emergence (or non-emergence) of E. huxleyi blooms, in dependence upon preexisting internal and external physical conditions, and b) an ability to predict this phenomenon dynamics in the future on the basis of climatic modeling.
This study has been supported by Russian Science Foundation under grant №17-17-01117.

Area A - Board 36

The arctic tundra is the terrestrial biome predicted to be most challenged by climate change [1]. Monitoring the tundra is therefore important for both climate scientists and decision makers. The Climate-ecological Observatory for Arctic Tundra (COAT) is an ecosystem-based, long-term, adaptive monitoring system created to provide this insight for Norwegian low- and high-arctic tundra regions [1].

One of the on-going challenges addressed by COAT is the ecosystem implications of large moth outbreaks in the ecotone birch forests bordering the low-arctic tundra. These outbreaks cause defoliation and tree mortality and can cause rapid state transitions from an initial healthy forest state towards states with strongly reduced tree cover [2]. The severity of moth outbreaks in the ecotone forest are believed intensified by northern range expansions of native moth species [3].

At the same time, higher temperatures are beneficial for the growth of trees and tall shrubs, and the encroachment of woody species into the tundra is expected as a long term result [1].

Analyzing the forest structure is therefore an important task for COAT, both for monitoring the state changes in the forest-tundra ecotone both due to warming, and due to the defoliation and forest mortality associated with moth outbreaks. The geographical areas to be monitored are fairly inaccessible, and remote sensing data is needed to cover the vast areas regularly.

The areas of interest have complete or partial cloud cover a significant portion of the time. While optical imagery can provide important information about foliage when it is available, Synthetic Aperture Radar (SAR) will be the primary data source as it can reliably provide data with frequent revisits.

Polarimetric SAR data can be decomposed into components related to different scattering mechanisms. These mechanism; double bounce, surface, and volume scattering, are useful for separating different vegetation types. Particularly the difference between double bounce scattering from stems and volume scattering from leaves can be useful for separating live from defoliated trees. There are several different methods to decompose the polarimetric data, such as Freeman-Durden [4], van Zyl [5], and Generalized Volume Scattering Model (GSVM) [6].

However, the tree density on the forest-tundra ecotone can be quite sparse, and all trees within a resolution cell are not necessarily in the same condition. Therefore using machine learning algorithms on the polarimetric decompositions can improve the vegetation structure mapping compared to simply looking at which scattering mechanism is dominant.

Selecting a suitable machine learning method is also an important aspect to consider. Supervised methods learn from labeled training data. Field work data for ground truth is expensive to obtain, and can only be used to label a very small number of pixels in a satellite scene. It is therefore crucial to be able to obtain good classification results even with little labeled data.

To deal with this, we consider different machine learning methods that can utilize the available ground truth data in the best way possible. For this application we are only interested in two classes; living and defoliated trees. The fact that we are not interested in segmenting other classes should influence our choice of algorithm.

One interesting approach is to use an ensemble of one-class classifiers using positive and unlabeled data [7].

Deep learning methods are the state-of-the art in machine learning, but generally require a large amount of labeled data. Pretraining deep networks on different datasets can mitigate this problem, but there is still a risk of over fitting when the number of training samples are low.

The area studied in this work is Polmak on the Norwegian-Finnish border in South Varanger (latitude 70.02°-70.04°, longitude 27.96°-28.01°). The study area stretches across the Norwegian-Finnish border. The effects on the forest from a major moth outbreak in the early 2000's is still clearly visible. This area is particularly interesting as the difference in reindeer herding regimes strongly affects how fast the birch forest recovers after a moth outbreak. The area on the Finnish side is used as feeding grounds for reindeer herds all year, while the area on the Norwegian side of the border is only used during winter.

Three fine resolution quad-polarization (FQ) C-band RADARSAT-2 scenes containing the target area were obtained. The nominal scene size is 25 km times 25 km with a resolution of 5.2 m times 7.6 m (range times azimuth). The acquisition dates are July 25th, July 31st, and August 1st 2017, and field work was conducted in the area between August 7th and 11th 2017.

As part of the fieldwork the vegetation was characterized for 165 points along six transects, each crossing the Norwegian-Finnish border. The distance between each point was 50 m. An area within a 5 m radius of the centre point was characterized with a particular focus on the presence and condition of trees in the plot.

We present preliminary results from this ongoing work on how to obtain the best possible classification map for living and defoliated forest states using different polarimetric decomposition models and machine learning algorithms. The resulting segmentation maps which show the extent of living and dead or defoliated forest states are of particular importance for monitoring state transitions in the forest-tundra ecotone as a result of biotic and climatic drivers.



[1] Rolf A. Ims, Jane Uhd Jepsen, Audun Stien, and Nigel G. Yoccoz, “Science plan for
COAT: Climate-ecological Observatory for Arctic Tundra,” Fram Centre Report
Series 1, 2013, Fram Centre, Tromsø.
[2] Ole Petter Laksforsmo Vindstad, Jane Uhd Jepsen, Malin Ek, Adam Pepi, and
Rolf Anker Ims, “Can novel pest outbreaks drive ecosystem transitions in northern-
boreal birch forest?,” Journal of Ecology, 2018.
[3] Jane Uhd Jepsen, Snorre B. Hagen, Rolf A. Ims, and Nigel G. Yoccoz, “Climate
change and outbreaks of the geometrids operophtera brumata and epirrita autum-
nata in subarctic birch forest: evidence of a recent outbreak range expansion,”
Journal of Animal Ecology, vol. 77, no. 2, pp. 257–264, 2008.
[4] Anthony Freeman and Stephen L Durden, “A three-component scattering model
for polarimetric sar data,” IEEE Transactions on Geoscience and Remote Sensing,
vol. 36, no. 3, pp. 963–973, 1998.
[5] Jakob J Van Zyl, Motofumi Arii, and Yunjin Kim, “Model-based decomposition
of polarimetric SAR covariance matrices constrained for nonnegative eigenvalues,”
IEEE Transactions on Geoscience and Remote Sensing, vol. 49, no. 9, pp. 3452–
3459, 2011.
[6] Oleg Antropov, Yrjö Rauste, and Tuomas Hame, “Volume scattering modeling in
PolSAR decompositions: Study of ALOS PALSAR data over boreal forest,” IEEE
Transactions on Geoscience and Remote Sensing, vol. 49, no. 10, pp. 3838–3848,
2011.
[7] Ran Liu, Wenkai Li, Xiaoping Liu, Xingcheng Lu, Tianhong Li, and Qinghua
Guo, “An ensemble of classifiers based on positive and unlabeled data in one-class
remote sensing classification,” IEEE Journal of Selected Topics in Applied Earth
Observations and Remote Sensing, vol. 11, no. 2, pp. 572–584, 2018.

Area A - Board 37

Keywords: ALS; LiDAR; change detection; Scots pine; forest growth
Forest site productivity, which is a quantitative estimate of the potential of a site to produce plant biomass, remains a fundamental variable in forestry and embraces two concepts: the site potential and that part of the site potential realized by a given forest stand. The most commonly used and widely accepted method of evaluating potential site productivity is the site index. Therefore, the construction of site index models describing stand height growth with age remains a fundamental task for site productivity differentiation. Site potential realized by a given forest stand may be expressed by current stand biomass increment. Three main data sources have been used for site index model development to date: 1) repeated measurements on permanent sample plots (PSP); 2) temporary sample plot (TSP) data from periodic inventories; and 3) stem analysis (SA) data. Stand biomass increment has been estimated to date using PSP and TSP plots. Our study is a practical application of change detection using airborne laser scanning (ALS) for the development of site index models and estimation stand biomass increment. We demonstrated how wall-to-wall multi-temporal ALS data obtained for large forest areas can be used in developing site index models that appropriately reflects site-specific growth trajectories. Repeated ALS data may be recognized as a new, fully valuable data source for TH growth, site index and stand biomass and biomass increment prediction. Repeated ALS observations can be a substitute for height growth data used in site index modelling and collected to-date from SA, PSP or TSP and enable a quite precise estimation of stand biomass increment. It could be expected that improving ALS technologies, decreasing costs of laser scanning acquisition and increasing data availability will result in improving the accuracy of forest growth estimates. Therefore, in the near future, both utility and increased predictive validity will lead to substantial increases in the importance of change detection using airborne laser scanning in forest growth prediction using the data from repeated ALS measurements.

Area A - Board 38

Biomass Retrieval Inter-comparison eXercise (BRIX) is a biomass retrieval algorithm inter-comparison exercise. It was conducted by European Space Agency (ESA) and multiple participants to develop algorithms for biomass estimation. The exercise aimed at using Synthetic Aperture Radar (SAR) P-Band datasets acquired as part of the ESA’s SAR campaigns in support of the upcoming ESA’s BIOMASS mission [1]. ESA set up the BIOMASS Algorithm Test Bed which is a virtual machine that (I) provides access to all BIOMASS campaign data in a unified format; (II) includes software tools that allow to implement and run algorithms in common open programming languages (Python, C, Fortran) as well as a limited number of proprietary programming languages (IDL, Matlab); and (III) makes available processing resources. Participants were invited to upload or develop their code, and run it on the Test Bed using the predefined campaign datasets.

BRIX consisted of two phases. In the first phase, four different retrieval algorithms were developed based on the data acquired during the AfriSAR campaign. One of the recent ESA’s SAR campaign that was conducted in Gabon in 2016. The SAR P band data acquired over four test sites: Lopé (super site), Mondah, Rabi and Mabounie. A complete set of data was made available to participants for Lopé including Tomographic SAR Stacks, Polarimetric SAR, Polarimetric Interferometric SAR, Digital Terrain Model (DTM), which were used as independent datasets in the models, and LiDAR derived Biomass, Canopy Height Model (CHM) and field measured data, used as reference data. In the second phase, the algorithms were assessed, on three other sites (Mondah, Rabi and Mabounie). There, only a limited set of data was available and no reference data was provided. ESA compared and evaluated the derived biomass maps over the three test sites with existing ground data and LiDAR derived biomass. The tested approaches include physically based models, empirical and machine learning techniques.

During the BRIX exercise following challenges were raised:
1. How to optimally use the available Polarimetric, Polarimetric Interferometric and Tomographic SAR data in one consistent retrieval concept.
2. The transferability of the retrieval developed over one site (Lopé) to three sites that are separated by several 100 km, which requires to account for different ecological characteristics and different forest structure.
3. How to correct the topographic effects inside SAR images.


REFERENCES
[1] T. Le Toan, S. Quegan, M. Davidson, H. Balzter, P. Paillou, K. Papathanassiou, S. Plummer, F. Rocca, S. Saatchi, H. Shugart and L. Ulander, “The BIOMASS Mission: Mapping global forest biomass to better understand the terrestrial carbon cycle”, Remote Sensing of Environment, Vol. 115, No. 11, pp. 2850-2860, June 2011.

Area A - Board 39

Remote sensing-based estimations of foliar nitrogen (N) concentration have gained a lot of attention in recent years. Foliar N is a primary regulator of physiological processes and an important indicator of photosynthetic and growth rates since it reflects the investment in photosynthetic enzymes, structure, and light-harvesting complexes of a plant. Both on leaf and on canopy level, empirical and physically-based models for N estimation using reflectance spectroscopy have been published. While they are able to reach quite high accuracies, a general problem of empirical model is their limited transferability.
We present results from several measurement campaigns including crown sampling, leaf spectroscopy, chemical laboratory analysis, and accompanying airborne imaging spectroscopy flights. On the leaf level, a partial least squares regression model was fit to a subset of the data and was able to explain over 90% of variance in the validation data set. When applied to new measurements outside the previous data range, the model was able to extrapolate and estimate the higher values without bias. This experiment was repeated with reduced spectral resolution. Although the validation data set of the original data could still be estimated with acceptable accuracy and without bias, spectra with over 50 nm bandwidth were not able anymore to extrapolate outside the known data range, making a strong case for high spectral resolution.
On canopy level we established regression models on imaging spectroscopy data and the reference measurements of foliar nitrogen concentration. We applied the model from the older data set to the newer one to see whether the extrapolation also works on image level. Furthermore, we applied regression coefficients that have been established on Northern American forests and published by the University of Wisconsin, Madison (Singh et al. 2015) to our European forest data sets and evaluated the transferability of these coefficients.

Singh A, Serbin SP, McNeil BE, Kingdon CC, Townsend PA (2015) Imaging spectroscopy algorithms for mapping canopy foliar chemical and morphological traits and their uncertainties. Ecological Applications 25:2180-2197 doi:10.1890/14-2098.1

Area A - Board 40

In this study, we investigate the relationship between the L-band vegetation optical depth (LVOD) retrieved by SMOS radiometer and forest attributes such as Above Ground Biomass (AGB) and Ecosystem Functional Properties (EFPs), in the framework of global monitoring of forest ecosystems changes.

AGB is one of the most important forest parameters, representing the amount of epigean carbon stocked. In the last decade, different studies exploited remote sensing and ground measurements to generate reference data and baseline maps of AGB, or carbon stocks, at continental, pantropical or biome level coverage (e.g. Baccini et al., 2008; Hu et al., 2016; Saatchi et al., 2011; Santoro et al., 2015). More recently, the work of Avitabile et al. (2016) combined two pantropical datasets and used additional field data to produce a new AGB map. Despite the use of improved methods, reducing the uncertainty in AGB upscaling, while providing updated information on AGB trends remains a challenge. Improvements in carbon mapping are expected with data from the forthcoming European Space Agency (ESA) ‘Biomass’ Earth Explorer, planned to be launched in 2021, or the recent Global Ecosystem Dynamics Investigation Lidar (GEDI) by National Aeronautics and Space Administration (NASA).
While AGB offers a static view of stored carbon, as a product of different plant processes, EFPs offer a dynamic view of carbon-related processes. Consequently, both aspects should be taken into account to effectively monitor forest status. EFPs characterize the variations in key processes, such as photosynthesis, respiration, or evapotranspiration, and their relationships, like water use efficiency of photosynthesis, that represents the amount of carbon gained per water transpired. Most EFPs describe the ability of vegetated ecosystems to optimize resources (water, light) for carbon uptake.
The SMOS satellite of ESA, launched in 2009, carries an L band radiometer on board. Although its main objective over land is to retrieve soil moisture, vegetation optical depth is also retrieved, and recent studies demonstrated that this variable contains information about forest height and forest biomass (Vittucci et al, 2018).

The present research investigates the potential of SMOS LVOD in providing relevant carbon-related data on tropical forests, considering both static and dynamic carbon information. The specific aims are to explore the links between SMOS LVOD and carbon processes, in Africa and South America homogeneous forests. SMOS nodes over forests were selected using the height database made available by Simard et al (2011). In particular, the conditions to have an average height above 5 m and at least 80% of SMOS pixel coverage with height above 5 m were applied. The study was conducted at both continental and ecoregions scales, since AGB and EFPs are known to vary at regional or biome scale according to abiotic climate variables, which in turn shape the distribution of biotic factors such as soil fertility, species composition and tree allometry. The ecoregions identified by Olson et al. (2001), which incorporate concepts from previous biome classifications, plus information on habitats and species assemblages, were also considered in the analysis.
In the first part of the study, the sensitivity of SMOS LVOD to biomass was tested for African and South American forests using two Random Forest models and the AGB map of Avitabile (2016) as reference data set. The ability to estimate AGB was evaluated with SMOS LVOD data for years included in the 2011-2015 period, also testing the improvement brought to AGB estimates by adding in the training set the climate variables from Climatic Research Unit Time Series (CRU TS) (4.01 release; Harris et al., 2014). At continental level, the estimated values of AGB using yearly averages of SMOS LVOD resulted to be significantly correlated with reference values of Avitabile et al (2016). The correlation coefficient R^2 is in range 0.7-0.8, for both continents. The analysis of LVOD yearly values in selected ecoregions from Olson et al. (2001) showed that, in general, LVOD decreases from wetter to drier regions. The coefficients of variation indicate a more stable signal in wetter forest ranges of both continents, such as moist, lowland and swamp forests, and higher signal dynamic in drier ranges.
The second part of the study is focused on the potential of SMOS LVOD in providing relevant carbon-related data through comparisons against EFPs data. For these aims, different reference datasets were employed, namely the EFPs generated in the framework of the on-going H2020 BACI project: Maximum Gross Primary Production (GPPmax), Bowen Ratio (BR), Light Use Efficiency (LUE), Water Use Efficiency (WUE), Rainfall Use Efficiency (RUE). The links between LVOD and the considered EFPs were explored for Africa and South America in different time intervals and still considering the ecoregion layers. First, the spatial relationship among EFPs and LVOD data was explored, by computing the Pearson correlation coefficients for data averaged at single year level and for the whole 2011-2014 period. The best results were obtained for LUE and WUE. Taking the 4 year averages, the Pearson coefficients for Africa were 0.86 for LUE and 0.84 for WUE, while in South America they were 0.72 for LUE and 0.70 for WUE. Then, temporal variations during the 48 months were investigated. Both EFPs and LVOD show significant seasonal variations. Large time shifts between the pairs of temporal trends are observed in drier regions, but are reduced in wetter regions.
Overall, the results of the study proved that LVOD can be usefully employed to estimate tropical forest biomass distribution at continental scale and to predict the trend of EFPs, advancing the previous efforts that established a correlation between LVOD dynamics and AGB. The availability of LVOD dense time series, can provide indicators for both static and dynamic forest attributes at global and regional scales.

Avitabile, V., Herold, M., Heuvelink, G., Lewis, S. L., Phillips, O. L., Asner, G. P., ... & Berry, N. J., 2016. An integrated pan‐tropical biomass map using multiple reference datasets. Global change biology, 22(4), 1406-1420.
Baccini, A., Laporte, N., Goetz, S. J., Sun, M., & Dong, H., 2008. A first map of tropical Africa's above-ground biomass derived from satellite imagery. Environmental Research Letters, 3.
Harris, I. P. D. J., Jones, P. D., Osborn, T. J., & Lister, D. H., 2014. Updated high‐resolution grids of monthly climatic observations–the CRU TS3. 10 Dataset. International Journal of Climatology, 34(3), 623-642.
Hu, T., Su, Y., Xue, B., Liu, J., Zhao, X., Fang, J., & Guo, Q., 2016. Mapping global forest aboveground biomass with spaceborne LiDAR, optical imagery, and forest inventory data. Remote Sensing, 8(7), 565.
Olson, D. M., Dinerstein, E., Wikramanayake, E. D., Burgess, N. D., Powell, G. V., Underwood, E. C., ... & Loucks, C. J., 2001. Terrestrial Ecoregions of the World: A New Map of Life on Earth: A new global map of terrestrial ecoregions provides an innovative tool for conserving biodiversity. BioScience, 51(11), 933-938.
Saatchi, S. S., Harris, N. L., Brown, S., Lefsky, M., Mitchard, E. T., Salas, W. , ... & Petrova, S., 2011. Benchmark map of forest carbon stocks in tropical regions across three continents. Proceedings of the National Academy of Sciences, 108(24), 9899-9904.
Santoro, M., Beaudoin, A., Beer, C., Cartus, O., Fransson, J. E., Hall, R. J. , ... & Thurner, M., 2015. Forest growing stock volume of the northern hemisphere: Spatially explicit estimates for 2010 derived from Envisat ASAR. Remote Sens. Environ., 168, 316-334.
Simard, M., Pinto, N., Fisher, J. B., & Baccini, A., 2011. Mapping forest canopy height globally with spaceborne lidar. Journal of Geophysical Research: Biogeosciences, 116(G4).
Vittucci, C., Ferrazzoli, P, Kerr, Y., Richaume, P., Vaglio Laurin, G., Guerriero, L., 2018. Analysis of vegetation optical depth and soil moisture retrieved by SMOS over tropical forests. IEEE Geoscience and Remote Sensing Letters, in press.

Area A - Board 41

As the biotic forest disturbances are an essential part of forests dynamics, the infestation of bark beetle decimates spruce forests for the last years not only in central Europe. Bark beetle (Ips typographus) plays an important role in the forests rejuvenation, hastens the decomposition of dead and dying wood, and may create early successional (snag) forest. Bark beetle attacks usually weakened trees, however, in case of its overabundance, attack also healthy trees. This status is called an outbreak and afflict central Europe for few last years. As sustainable resource management (e.g. maintaining tree species heterogeneity) is the most powerful way of defence, we need to make the massive prevention of bark beetle outbreaks. When the outbreak appears, selective and even square logging used to be applied. The most crucial step for reducing damages and losses is to find and clean up infected trees before the evolution of bark beetle generation is finished. Finding infected trees across the forest in quite demanding and time-consuming process even for a skilled forester. That is why the finding of solutions for the detection of infected trees using remote sensing techniques is still a current research topic.

The Czech Republic deals with outbreaks of the spruce bark beetle species due to many even-aged spruce forests. The aim of our study is to find a solution for detection of infected trees, which are in “green attack stage”, i.e. the trees are infected, but any visual sign appeared at the tree so far. This phase is the most important to be detected and trees identified for treating action. For early detection of infected trees across larger areas, Unmanned Aerial Vehicles (UAVs) equipped with proper sensors seems to be a cost-effective solution. In our study, we did a low altitude aerial survey using a consumer-grade RGB camera and a modified Near-Infrared RGB camera (NIR) in June, as well as the first and second half of August, and in October. Using an object-based image analysis and clear statistical analysis, we found predictors to compute for early detection of infected trees. Results show that even using low-cost (modified) cameras it is possible to detect early attack stage of bark beetle infection on an individual tree level. We conclude simple proportional vegetation indices that take advantages of red, green and near infra-red bands, as the best predictors at early stages of infection.

Area A - Board 42

Forests cover about 30 percent of the earth surface, they are the most biodiverse terrestrial ecosystems and they are at the base of many ecological processes and services. The loss of forest biodiversity makes in risk the benefits that the humans derived from them. The assessment of biodiversity is therefore an important and essential goal to achieve, that however can result difficult, time consuming and expensive if estimated through field data. Through remote sensing, it is possible to estimate in a more objectively way the species diversity, using limited resources, covering broad surfaces with high quality and standardized data.
Forests cover about 30% of the Earth surface, they are among the most biodiverse
terrestrial ecosystems and represent the bulk of many ecological processes and services. The loss of forest biodiversity makes in risk the benefits that the humans derived from them. The assessment of biodiversity is therefore an important and essential goal to achieve, that however can result difficult, time consuming and expensive if estimated through field data. Remote sensing covers large areas and provides consistent quality and standardized data which might prove useful to estimate species diversity. One of the methods to estimate biodiversity from remote sensing data is through the use of optical data, applying the concept of the Spectral Variation Hypothesis (SVH) , which states that, the higher the spectral variation of an image, the higher the environmental heterogeneity and the species diversity of that area. The SVH has been tested in several ecosystems, with different optical data, using different indexes and measures; recently in literature, the Rao’s Q index, applied to remote sensing data has been theoretically tested as a new and innovative spectral variation measure. In the first part of this study for the first time, the SVH has been tested in a forest ecosystem, through the Rao’s Q index and through other two well know heterogeneity indices (coefficient of variation and Shannon’s H, used as benchmark).
We tested the SVH with an NDVI time series from 2016 and 2017, derived from the Sentinel-2 (with a spatial resolution of 10m) and Landsat 8 satellites (spatial resolution of 30m) to understand how the hypothesis changes during and through the years. The Spectral variation of the images were correlated with data of tree species diversity (through Shannon’s H) collected in 20 forest plots (mainly covered by conifers) of 1ha each.

The second part of this research is focused on the use of the LiDAR data to better understand the vertical structure of the forest that, according to different studies, is related to the species diversity of the forest itself. A new concept called Height Variation Hypothesis (HVH) has been devised and developed in this study, stating that, the higher is the variation of the height of the trees (Canopy Height –CH-) derived from LiDAR, the more diverse is the structure of the forest and the higher is the species diversity of that area. In this case, the concept of HVH has been tested through the Rao’s Q index and correlated with data of tree species diversity (through Shannon’s H) collected in the mentioned 20 forest plots.

The results showed that the Rao’s Q is a grateful variation index for the study of the SVH and for the HVH. In the study of the SVH, for both the sensors, the correlation with the field data had the same tendency as the NDVI trend, reaching the highest value of correlation (through the coefficient of determination R²) in June, when the NDVI was at its peak. In this case the correlations reached a value of R²=0.6170 (2017) and R²=0.48 (2016) for the Sentinel-2 and of R²=0.4542 (2017) and R²=0.47 (2016) for the Landsat 8, showing that the SVH is scale and sensor dependent. The CV had also the same NDVI tendency, reaching in average highest values in June but anyway lower than the Rao’s Q. The Shannon’s H showed in general, for both the years, low level of correlation.
The HVH showed also excellent results: the correlation with field data reached values of R²=0.74 with CH with spatial resolution (SR) of 2.5m, showing a decrease of correlation with the decrease of the resolution (SR 5m R²=0.73; SR 10m R²=0.62; SR 20m R²= 0.15) underling that also the HVH is scale dependent.
The study of forest biodiversity through the concept of the SVH and HVH using the Rao’s Q index showed grateful and promising results in the studied forest and could lead to as much good results with other remote sensing data or in other forests or ecosystems.

Area A - Board 43

Forest disturbance are inherent ecosystem processes and the disturbance regime of an ecosystem is described by the size, shape, frequency and severity of disturbance events occurring in this particular ecosystem. There is accumulating evidence that climate change and land use alter disturbance properties globally, potentially shifting disturbance regimes beyond their safe operating space, with unknown impacts on ecosystem functions and services we as humanity rely on. Assessing changing disturbance regimes is challenging, though, as it requires (i) a spatially explicit but consistent assessment of forest disturbances that (ii) covers long-enough time periods to derive regime indicators. We tackle this challenge by mapping (changes in) European forest disturbances from the Landsat archive. Europe is characterized by a high diversity in ecosystems and land use practices, highlighting the need for a spatially consistent assessment across administrative and biogeographic regions. To achieve our goal, we first used a sample-based visual assessment of > 20,000 Landsat time series to estimate changes in disturbance rates across Europe. Second, we used the sample-based information to train models mapping spatial pattern of forest disturbances using state-of-the-art time series algorithms (i.e., LandTrendr ensembles). Third, we developed a set of disturbance regime indicators that are explicitly suitable for characterizing disturbance regimes from Landsat based disturbance maps. Results from Central Europe show that forest disturbance rates have doubles from 0.5 % per yr. to 1 % per yr. in the last three decades. However, increases were strongly driven by non-stand-replacing disturbances, documenting a shift from clear-cut based silvicultural systems to “close-to-nature” systems and retention forestry. The disturbance maps (overall accuracy of 80 % with balanced commission/omission errors) showed that while temporal trends were largely consistent across space, spatial-temporal disturbance patterns varied substantially. Based on this variability, we could map distinct disturbance regimes that reflect both land use (history) and biogeographic variability in tree species and disturbance agents. We conclude that Landsat time series are well suitable for mapping disturbance regimes across diverse systems such as Europe. Those maps will serve as important baseline for assessing changes in disturbance regimes under climate change. Further, we highlight that Europe’s forests are complex human-environmental systems, where both natural and human factors shape current and future disturbance regimes.

Area A - Board 44

Uncertainty in describing processes that control carbon outputs and their cycling rates in terrestrial ecosystems is a major contributor to overall uncertainty in Earth System Models. Here, we focus on remote sensing approaches to understand the status and dynamics of above-ground carbon related to vegetation as primary producers in the tundra-permafrost coupled system of the central Lena Delta in Subarctic Siberia.
Historically, spaceborne and airborne remote sensing approaches focus on classifying land cover depending on the spatial resolution of the sensors. Developments in optical remote sensing are ongoing for quantification of vegetation-related metrics and by this, monitoring of „spectral plant traits“ in space and time including above-ground carbon dynamics.
In ongoing and future Russian-German cooperation on Lena Delta expeditions and with the opportunity of the modern Samoylov research station and new spaceborne satellite missions we plan to evaluate the applicability of remote sensing for assessing vegetation stocks and short-term fluxes in the Lena River Delta.
i) Synoptic mapping of above-ground vegetation biomass for the Lena Delta region:
We plan to upscale vegetation cover plot information and in-situ biomass estimates from harvesting to the plant traits 'above-ground biomass' and successively to above ground carbon. We will use Landsat-8 OLI and Sentinel-2 sensors for the spatial upscaling providing mapped fields of biomass (and above-ground carbon as higher level product) on 6 m x 6 m (RapidEye), 10 m x 10 m (Sentinel-2) and 30 m x 30 m (LANDSAT) grid cell spatial resolution.

ii) Synoptic monitoring of vegetative regrowth offsetting losses of carbon by spring floods and thermo-erosion.
We plan to use
-ground-based and unmanned aerial vehicle (UAV) based data: geobotanical, dendrochronological and sedimentological assessments to retrieve vegetation community, biomass, pool ages of vegetation communities, time and type of disturbance.
-satellite-based upscaling and monitoring on different spatial scales and on short (Landsat 8 OLI, Sentinel-2, RapidEye) time scales (i.e. 2 to 5 years depending on sensor).
In doing so, we contribute to the understanding of variability of carbon fluxes dependent upon vegetation community, biomass, time and type of disturbance and stand age.

iii) Photosynthesis is the starting point of the carbon cycle. Can we assess landscape-scale inherent dynamics of photosynthetic activity for the central Lena Delta region using remote sensing?
Remote Sensing vegetation indices, most commonly the Normalised Differentiation Vegetation Index, NDVI, are widely used to model the fraction of photosynthetically active radiation absorbed by vegetation, fPAR, for estimating gross primary productivity, GPP, from remotely sensed data. We will investigate the applicability of MODIS and SENTINEL-3 derived NDVI (offering high temporal resolution) for the tundra landscape of the Lena Delta region. We also will investigate the usability of the Sun-induced Fluorescence SIF remote sensing product (produced at the GFZ) as an indicator of photosynthetic activity. Assessments of the applicability of the remote sensing products will be done using upscaling: UAV-based imagery, higher spatial resolution Rapid Eye and assessing the leaf development, plant shooting, leaf broadening and greening by phenological monitoring from time lapse cameras.

Acknowledgments. Ground-based measurements have been supported by the Russian-German LENA expeditions and the BMBF-funded project „KoPf”

Area A - Board 45

Vegetation structure and biochemical parameters, including LAI, fCOVER, fAPAR, leaf chlorophyll concentration and leaf water concentration, are widely used to scale ecosystem models in space and time. While many of these parameters have been derived using multispectral imagers there is substantial evidence that the inverse problem of estimating them from directional reflectance measurements is ill-posed. Here we provide a new approach for simultaneous retrieving both a structural parameter (the directional area scattering factor) and a biochemical parameter (chlorophyll concentration) using Sentinel-2 multispectral measurements without prior constraints. These retrievals are then used to constrain inversions of other parameters using the SL2P processor.

Results comparing the new approach with baseline SL2P retrievals and field measurements are presented over a range of crop and forested sites in Canada and within the VALSE2 project over Europe.

Area A - Board 46

TTo better understand the functional vegetation dynamics from remote sensing, the spectral characterization of the dynamic photoprotection mechanisms are key. In addition to energy dissipated by sun-induced fluorescence (F), photoprotection or non-photochemical energy quenching (NPQ) controls the electron transport to the photosystem reaction centres. Both on the diurnal as well on the seasonal scale the NPQ mechanisms have an impact of the vegetation dynamic reflectance spectrum. However, the in-depth identification and description of specific spectral contiguous signatures linked to the various NPQ dynamics are still poorly understood, especially from a proximal or remote sensing point of view. Photoprotection dynamics from remote sensing are commonly measured with the Photochemical Reflectance Index (PRI), but this index is prone to several dynamics in the short diurnal term. Distinguishing and associating specific spectral signatures with the NPQ processes under varying environmental conditions would significantly advance the knowledge on quantification of NPQ and its different components. Passive optical spectroscopy dedicated to vegetation reflectance, transmittance and F measurements has a great potential to study such functional leaf dynamics.

When exposed to excessive light, several controlled leaf photoprotection mechanisms can be activated. We studied spectrally contiguous (hyperspectral) transient absorbance and passive F dynamics of leaves in the visible and near-infrared wavelengths (VIS-NIR, 400-800 nm) after sudden strong natural illumination exposure. For several species we found major dynamic absorbance events with a different onset and kinetical behaviour. Besides the absorbance feature related to the chemical xanthophyll conversion, several additional overlapping NPQ absorbance features are present in the vegetation spectrum. Slowly induced but strong conformational changes, related to the pigment-protein turnover, appear to be responsible for a significant absorbance change across the full VIS-NIR range. A green peak around 550 nm and a wide near-infrared peak were observed. This strong absorbance increase shows the in vivo VIS-NIR slow conformational change, a strong photoprotection mechanism characterized by an energy redistribution and promotion of lower energy state transitions across the pigment bed. The characterization of these hyperspectral absorbance features is a further step in the understanding of dynamic plant strategies in nature regarding its energy distribution, i.e. with a harvesting or a dissipation fate, and a promising direction in the quantification of dynamic photoprotection from hyperspectral sensors and future EO missions.

Area A - Board 47

The leaf area index (LAI) is an important biophysical indicator of productivity, photosynthetic capacity, and degree of stress, closely related to reflectance of vegetation and various biosphere-atmosphere interactions. LAI is one of the essential climate variables defined by the Global Climate Observing System (2010), and it is often used to assess ecosystem services and characterize the structure and functioning of forest ecosystems.
Remotely sensed data have been widely used in quantifying biophysical properties of different vegetation types. This study describes the retrieval of forest biophysical variable of LAI from Sentinel-2 MSI level-2A imagery in a heterogeneous Mediterranean environment. The study area is located in Northern Pindos National Park (north-western Greece) known for rich biodiversity, rich fauna and natural vegetation, and its wild and imposing beauty. The field collected data contain biophysical parameters measurements of a total 60 Elementary Sampling Units (ESU) among different forest types.
Within this framework, Gaussian processes regression (GPR) algorithm was used to model the relationship between field biophysical-LAI measurements and spectral data. GPR has been evaluated as a promising nonparametric regression technique in terms of processing speed and accuracy. GPR, as Bayesian state-of-the-art tools for discriminative machine learning and dimensionality reduction, provides insight in model features carrying relevant information and also in theoretical uncertainty estimates.
Since LAI has been strongly related to spectral measurements, vegetation indices (VIs) are among the oldest and most widely used tools to estimate LAI. Most of VIs rely on two-band formulations reducing multispectral data to a single variable for predicting and assessing vegetation characteristics. A typical problem of two-band spectral indices is that they rapidly lead to saturated values of the biophysical property in question. This can be partly overcome by optimizing band selection, alternatively, three-band spectral indices (3BSIs) have been proposed to avoid the saturation problem in many two-band spectral indices.
Specifically, in this study we considered VIs included two-band and three-band index formulations - simple spectral bands ratios, traditional spectral indices and red-edge and near-infrared shoulder-based indices - of the Sentinel-2 MSI imagery, and we specify the optimal VI for LAI estimation. Furthermore, the performance of single bands was evaluated, and the optimal number of bands was assessed by using the backward elimination approach.
Overall, the objective of our study was (i) to evaluate the capability of GPR assessing LAI over heterogeneous Mediterranean forest using Sentinel 2 MSI imagery (ii) to determine the optimal VI and spectral band for LAI estimation ii) to generate maps of LAI, based on the best evaluated approach, (iii) to compare the results with LAI values obtained from the Biophysical Processor installed on SNAP Toolboxes.

Area A - Board 48

Plant Traits (PTs) are able to provide crucial information to understand the ecosystem functional properties and the mechanisms underlying the provision of ecosystem services. Many Spectral Vegetation Indices (SVIs) have been adopted to retrieve PTs, and the so-called red-edge indices (combining bands on the borderline between near-infrared (NIR) and visible (Vis) bands were shown to be particularly good proxies of pigments absorption-related traits such as chlorophyll content, LAI, and Phytomass.
In this work, we aimed at exploring the use of indices solely based on NIR bands (wavelengths >740 nm), which are mostly greenness-independent, and are expected to be sensitive to canopy architecture information. The link between canopy structural traits and NIR reflectance was investigated, and the use of NIR spectral information to retrieve canopy structural information on both temporal as well as on a spatial basis was tested. Ground measurements of hyperspectral reflectance and phytomass were carried out at three grassland sites (Monte Bondone, Italy; Neustift, Austria; Majadas de Tietar, Spain) using a hand-held spectroradiometer. The grasslands show a very wide range of LAI and an extreme range in species composition.
From the preliminary results obtained by field observations and RTM (Radiative Transfer Modeling), it emerged that the relationships between NIDI (Near Infrared Difference Index) indices and structural plant traits (phytomass, LAI) on temporal basis show very high correlations coefficients (R2= 0.85) compared to “traditional indices such as NDVI, CI-red edge and MTCI. Considering the very promising results we obtained on temporal basis, we carried out further investigations on a spatial basis. On the spatial basis we realized that there were much poorer correlations between NIDI indices and PTs. The possible explanation of such poor correlation is that when many structural PTs co-vary (e.g. in spatial observations, when canopies show very different canopy architecture) SVIs and structural PTs relationships may be complex and quantitative PT estimations using spectral data may be difficult.
Further research was aimed to adopting a RTM approach to investigate more in detail the link between structural traits and the NIR reflectance response and we identified the key plant traits (structural, chl and water content) on the NIR spectral response, and in particular on NIDI indices based on Sentinel-2 simulated bands. The RTM results seem to indicate that when one trait is changing and the other are constant (this can happen when we carry out temporal observations), we are able to estimate structural PTs (This can happen when we carry out temporal observations). Further study will be focused on adopting Sentinel 2 data to retrieve grassland PTs.

Area A - Board 49

Vegetation, through its condition, reflects properties of the environment. A variety of alpine plant characteristics have be analyzed using remote sensing tools. The study aimed to assess the condition of reculivated alpine swards dominant species (Agrostis rupestris, Festuca picta and Luzula alpino-pilosa) using hyperspectral remote sensing techniques. The study area was the UNESCO Man&Biosphere Tatra National Park (TPN, Poland), in particular the Kasprowy Peak, surroundings hills and the Red Peaks of the High Tatras where the high mountain grasslands are strongly influenced by tourists.
The data were collected during the field studies carried out from 2012 to 2014. All the polygons were broken down to: (a) areas subject to trampling, which are located in the neighborhood of trails, at sites clearly destroyed by tourists; (b) reference areas with similar habitat conditions located near to trampled areas but normally at a distance of 10 m from trampled polygons; (c) remediation areas, i.e. patches delineated by the TPN with limited access, e.g. due to the application of the so-called fladry or jute mats. The following measurements were made in each of the tested polygons: (I) spectral properties using an ASD FieldSpec 3 spectrometer using the ASD PlantProbe; (II) leaf chlorophyll content measurements using a Chlorophyll Content Meter CCM-200 that measures the Chlorophyll Content Index (CCI); (III) chlorophyll fluorescence using the Plant Stress Meter fluorometer (PSM Mark II, Biomonitor) and LeafClip applied to create dark adaptation that terminates the photosynthesis processes. Measurements were also performed without dark adaptation; (IV) air thermodynamic temperature (ta) and leaf radiation temperature (ts) using the IRtec MiniRay 100 with an insert probe pyrometer in order to evaluate the water stress using the thermal infrared (TIR); (V) measuring energy balance of Absorbed Photosynthetically Active Radiation (APAR) and the fraction of APAR using a line ceptometer AccuPAR model 80.
The obtained spectral properties (for trampled species, reference and remediation areas) were statistically significantly in terms of cellular structures, chlorophyll and water content in the canopy. The greatest differences were for the spectrum range describing the absorption of photosynthetically active pigments (448-514, 581-707 nm), cell structures (1385-1556 nm), and water content (1801-1835, 1845-1862, 1879-2500 nm). Poorer index values were observed for trampled plants, especially for chlorophyll based indices (e.g. ratio analysis of reflectance spectra algorithm chlorophyll a), general condition (e.g. normalized multi-band drought index) and water content (e.g. water band index).
In the mountains, increased light intensity may be observed, and so the Fv/Fm rates fluctuate between 0.471 and 0.810. The median rates for Fv/Fm (in the case of reference and remediated plants) falls within the range of 0.700-0.755, whereas for species subject to trampling has 0.500-0.655 as the respective range. The species of alpine swards growing on reference areas were marked by a high correlation between Fv/Fm and PRI: Festuca picta (Rs=0.65). The rates of Fv/Fm' demonstrated the highest correlation with PRI (Rs=0.40 for Festuca picta from trampled areas; Rs=0.60 for Festuca picta in reference areas); and CAI (Rs=-0.51 Agrostis rupestris from trampled areas).
Species in remediation areas were characterized by a similar, or sometimes even better, properties than the reference areas that was also confirmed by fluorescence measurements (Fv/Fm). In general, remote sensing indices were within optimal ranges, so the condition of the high grasslands was determined as good. Overall, the fluorescence analysis and remote sensing tools confirmed the suitability of such methods for monitoring species in remote mountain areas.

Area A - Board 50

Solar induced fluorescence (SIF) is explored as a novel remote sensing signal that marked for its potential to be considered as a direct indicator of plant functional activity and health condition. The present state of the art of SIF is extremely relevant and highly attractive field of research for scientific communities and space agencies such as ESA, NASA due to its ability to quantify plant photosynthetic process. SIF signal emitted by plants is a key component to understand the activities of various terrestrial ecosystems and its respective vegetation groups more deeply.
Peatlands are one of the most significant terrestrial ecosystems in the earth that play a vital role in the global carbon cycle. It stores approximately one-third of the world’s soil carbon and in any change in the carbon balance directly impact the global carbon content. Due to huge heterogeneity, internal diversity, steep environmental gradients, high spatial and spectral variability, it is always a challenge to predict and simulate the SIF signals through commonly used reflectance based vegetation indices (VIs) for heterogeneous ecosystems such as peatland.
This research will show for the first time, how narrow-band hyperspectral imaging technology derived VIs and simple fuzzy approach can strongly predict the red (O₂B) and far-red (O₂A) SIF signals from heterogeneous ecosystems of peatland, grasslands and forests. Raster-based fuzzy model attributed with fuzzy membership functions have been considered to simulate the SIF from diverse vegetation groups and further was validated by the measured SIF signals. The study was conducted in and around Rzecin (POLWET) peatland located in the central-western Poland (Wielkopolska region) within Noteć Forest (52°45′N latitude,16°18′E longitude, 54 m asl). The entire Rzecin peatland and its surroundings are enriched with diverse vegetation groups from Pinus sylvestris dominated forests, riparian forest, to rush peatland vegetation dominated by Phragmites spp., Typha spp. and Carex spp. and with different species of Sphagnum spp. The study incorporated airborne imaging spectroscopic data of ‘HyPlant’ acquired during the SWAMP campaign funded by the COST Action OPTIMISE (ES1309), EUFAR (FP7) and ESA under the FLEX-EU mission framework and held on 11th July 2015 over the Rzecin peatland. A combination of peatland vegetation groups, natural and managed ecosystems such as mowed meadows, deforested land, grassland, post-agricultural land, riparian forest, rush vegetation, secondary pine forest community, semi-natural pine forest, transition mires, and young deciduous forest located in and around Rzecin peatland have been examined for this study. All the patches were considered from authenticated and detailed botanical survey data from 2015.
‘HyPlant’ DUAL module derived Simple Ratio (SR), Normalized Difference Vegetation Index (NDVI), Red-edge Normalized Difference Vegetation Index (NDVIre), Enhanced Vegetation Index (EVI), and Photochemical Reflectance Index (PRI) have been used as model parameters and processed through justified fuzzy membership functions. Based on literature supported justifications to SR, NDVI, NDVIre and EVI indices, which are highly sensitive to SIF, we received high memberships in proportional trend, whereas for PRI we received high membership in inversely proportional trend. Furthermore, fuzzy overlay operation has been employed to simulate the SIF signals. Finally, the simulated SIF values of different vegetation groups had been validated through Spectral Fitting Method (SFM) retrieved SIF signals.
The result shows statistically significant (p<0.00) and good agreement between fuzzy simulated SIF values with established O₂A and O₂B SIF signals with the R2 = 0.81, and R2 = 0.90 respectively. To understand the model strength to predict and simulate SIF, in the next step different combinations of 3 {f(NDVI+EVI+PRI)} and 2 {f(NDVI+PRI)} indices have been processed further. Outcomes of our analyses confirm the model strength for SIF prediction, supported by statistically significant and strong correlation between simulated data and established SIF signals. The high confidence of prediction through fuzzy model employing simple reflectance based VIs is proved to be advantageous in data constraint situations and can also act as a strong proxy for SIF. It can further assume that the model can work as a potential indicator for plant photosynthetic efficiency and health conditions and fill the limitations of SIF data.

Keywords: Sun-induced fluorescence, Vegetation indices, Fuzzy model, HyPlant, Peatland

Acknowledgment
The Research was co-funded by the COST Action ES1309 OPTIMISE, FP7 European Facility for Airborne Research (EUFAR), FLEX-EU ESA (Contract No. 4000107143/12/NL/FF/If CCN3) and European Space Agency which supported the SWAMP training course (6-16th of July 2015) and airborne campaign on 11th of July 2015. Analyses of the data were supported by the Polish National Research Centre within the project No 2016/21/B/ST10/02271 (Sun Induced florescence and photosynthesis of peatland vegetation response to stress caused by water deficits and increased temperature under conditions of climate manipulation experiment).

Area A - Board 51

Spectroscopy captures vegetation biophysical characteristics associated with photosynthetic function, which vary diurnally and/or seasonally. Field, airborne and satellite spectroscopy show great potential for measuring remotely vegetation bio-physical and functional characteristics.
Consistent and calibrated spectroscopy observations are required at all temporal and spatial scales, to bridge the gap between in-situ leaf, proximal, air- and space-borne observations. New generation spectral systems are needed to reliably capture the diurnal and seasonal dynamics in vegetation function, under varying solar illumination conditions.
Coordinated research efforts are underway in the US and EU, utilizing homogenized high-frequency time series, to develop canopy Chlorophyll content maps, and assess the seasonal changes in chlorophyll and productivity for key ecosystems.
We will present our current results and findings from the use of the homogenized satellite data-sets and two filed campaigns in 2017 and 2018, and will discuss current strategies to detect and monitor the essential changes in plant traits associated with vegetation function.
This investigation aims to contribute for bridging the gap in Earth observation between field and airborne measurements. Our first objective is to produce consistent medium and high resolution canopy chlorophyll product prototypes, and robust algorithms, that can reliably be validated and scaled to regional and continental scales. The key objectives of the work are to: 1) use in a seamless fashion the dense remote sensing time series to develop algorithms for estimating canopy traits (e.g. chlorophyll content); 2) contribute to better understand the effects of the diurnal and seasonal dynamics in environmental conditions on vegetation traits, metabolic processes and productivity.

Area A - Board 52

The undergoing climate changes are impacting both the environment and human welfare. Temperature is expected to reach higher values in every season, generating an annual growth. The Danube River Basin, and particularly the South-East area, that includes most part of the Pannonian Basin, requires a particular attention as the main future trends suggest that the highest temperature increases and the amount of precipitation decrease, while extreme events (e.g. torrential precipitations, widespread droughts) are more likely to appear. One of the main scientific challenges identified in this region is the enhanced characterization of droughts.

Over the last decade, rapid developments in digital technologies and in our capability to monitor Earth from space with EO satellites have led to new and huge opportunities for science and society. Drought monitoring is a challenging use case to exploit the potential of EO data archives and the ongoing field development. Europe is leading this sensing revolution in space through recent European initiatives like: Copernicus, ESA Climate Change Initiative and the corresponding development of a family of Sentinel missions, enabling global monitoring of our planet across the whole electromagnetic spectrum on an operational and sustained basis.

Artificial Intelligence (AI) is an important instrument enabling exploration of big data and bringing new insight and predictive capabilities particularly when used together with physical principles and scientific interpretation. AI is expected to unlock the full potential of EO data streams to better understand the changes of our planet and predict its evolution. Although not widely explored until recently, AI became recently a revolution and started to emerge due to numerous initiatives worldwide, including in the field of EO.

The paper presents the development of a set of innovative EO-based products and enhanced data sets dedicated to describe drought processes in the Pannonian Basin. The methodology is exploiting the potential of EO open data jointly with the existing in-situ sensor network across the Pannonian Basin and numerical models to produce comprehensive data sets for drought monitoring and forecast. EO data collection include MODIS and Landsat data to analyze and interpret long time series of drought events, while Sentinel-2 and Sentinel-3 are used to derive products for near-real time drought monitoring. Deep-learning (DL) algorithms, based on Deep Neural Networks, are used to improve the performance of traditional image processing methods.

Area A - Board 53

Presently, the Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) is a widely used index for the study of plant cover. It allows estimating the seasonal and long-term dynamics, since FAPAR-based data analysis is available for multiple satellite sensors.

Volga-Akhtuba floodplain (VAF) wetland ecosystems are very sensitive to Volga river runoff, and particularly to spring inundation, which currently depends on Volzhskaya Hydropower plant (HPP) downstream water discharge regulation, as well as on the water natural inflow into the upstream reservoir. The plant communities in VAF are extremely heterogeneous, both spatially and temporally. We previously showed that different types of plant community have its unique features in seasonal vegetation index dynamics, which may have one, two or three local vegetation waves, clearly reflecting the actual in-situ dynamics.

The present study of VAF plant ecosystem dynamics focuses on the assessment of vegetation indicators using SeaWiFS and MERIS FAPAR data. The specific FAPAR curves, model functions, and its parameters for 12 test sites represent the most common plant community types in VAF. We compare the period of relatively high Volga river runoff (1998–2005) with the period of relatively low runoff (2007–2011), with the extremely dry 2006 in between them.

Our quantitative analysis of vegetation activity bases on a harmonic approximation of FAPAR time series. Instead of discrete data points, it yields a continuous function suitable for further analysis as a model yearly FAPAR curve. This methodology suggests a simple, clear and fairly universal approach to analyzing FAPAR time series.

The analysis shows that in most test sites the model indicators changed considerably in 2007–2011, and 2006, as compared to the period of 1998–2005, due to the changes of the Volga river runoff, indicating significant changes in ecosystem structure and species composition. The in-situ observations also confirm these facts. In 2006–2011, the transformation of water meadows into secondary steppe or weed communities proved to be not a local processes, but occurring all over the VAF, being most critical in its upstream segment near the HPP.

As follows from our study, the method provides weighted and unbiased quantitative analysis of plant ecosystem dynamics. It also gives a clear estimation of ecosystem health and allows detecting changes in its type for a given territory, e.g. arid wetland degradation under the unfavorable environmental conditions and human impact in our case.

Area A - Board 54

One of the main questions related to ecosystem functioning at seasonal and interannual time scales with primary value for socio-economic applications is vegetation productivity, whether of crops or forests and its fluctuation. Given the geographical coverage offered by the MSG satellites, the paper illustrates the possibility of providing information concerning food early warning systems.
This work is based on a cybernetic insight in ecosystem functioning realized through modelling of energy flow according to the first two principles of thermodynamics. Thermodynamically the ecosystem is a steady state open system, energy enters and stored or degraded and thus their functioning may be expressed by construction of the entropy balance of the system. The balance between information gained and entropy generated in a unit of times determines the dynamic state of the ecosystem. Methodologically, this concept is based on the thermodynamic description of energy flow conversions through the ecosystems and the established on these basis relations between biogeophysical and biogeochemical cycles.
The quantitative description of energy flow in agroecosystems is performed according Surface-boundary approach in biothermodynamics. The energetic losses accompanying ecosystem functioning are quantified to serve as a measure of specific ecosystem-site capacity to utilize the solar energy into biochemical energy and related growth response/crop yield. As such a complex thermodynamic quantity for energetic climatic resources, the entropy production (in terms of thermodynamics - function of dissipation, Ф) in the main photosynthetic apparatus (i.e. canopy leaves) is used. Ф is defined as a sum of the entropy produced during the heat- and the water- exchange between canopy leaves and environment.
Applying this thermodynamic modeling approach to agricultural field, climatic mean of Фfor territory of Bulgaria is quantified. Analytical description of the functional relation between country yield (winter wheat) and Ф is obtained, and used for crop yield prediction in three categories (high/moderate/low). In parallel, MSG based information is used for quantification of Ф-index using ten-years mean of the difference between skin- (LSA SAF LST) and air temperatures as well as of evapotranspiration flux (LSA SAF ET). It is illustrated that data from meteorological geostationary satellites can be used for bioenergetic climatic resources - yield forecast through the same thermodynamic model.
Regarding global forecasting applications, in-situ observations cannot fulfil alone the modelling needs. Here we highlight how satellites data are used for modeling and diagnoses of agricultural component of biosphere functioning at local and regional scale for Bulgatia. This approach is applicable at any other region accounting for specific crop, climate, soil environment.

Area A - Board 55

Lightweight drones (also referred to as unmanned aerial vehicles, UAVs) have been shown by others to provide a cost effective method for acquiring finely resolved spatial data. In recent years, they have been applied extensively to deliver information to scientists in a range of disciplines, including in the study of vegetation canopy traits. To date, different kinds of drone-based data (i.e. point clouds generated from Structure from Motion photogrammetry (SfM) and indices derived from multi- or hyperspectral acquisitions) have largely been used independently to estimate canopy traits. Furthermore, there is a lack of studies attempting to integrate the fine grained spatial and structural information from drones into workflows relying on coarser resolution airborne and satellite data in a physically-meaningful way. This presentation will explore the extent to which this integration is possible, using drone-data based workflows in a simple study system.
Data were acquired in Summer 2018 during a collaborative field campaign initiated by ESA over a corn field near Braccagni, IT. The drone mounted sensors used were an RGB camera (Ricoh GR II) and a multi-camera array possessing green, red, red-edge and near-infrared bands (Parrot Sequoia). Reflectance panels used for empirical line correction and assessment of the multispectral drone data as well as optical properties of the soil and corn leaves were measured using field spectrometers (ASD FieldSpec and Spectral Evolution SR-3500). Additionally, airborne hyperspectral data (HyPLANT) were also acquired concurrently with the drone flights as well as a cloud-free acquisition from the Sentinel-2 A satellite one day later. Leaf area index (LAI), chlorophyll content related values (using a SPAD-502 instrument) and canopy height values were measured in multiple 10 x 10 m plots.
Drone data quality was rigorously assessed with respect to independent spectral validation observations of reference panels as well as soil and grass surfaces (point-based and airborne). With reference to the Parrot Sequoia sensor, there was a need, as others have demonstrated, to address issues with the radiometric correction of the data, when applying the empirical line method. Fine grained height information was derived from drone SfM photogrammetry products and compared to field validation values (field measurements with decimetre precision), and used to separate the corn canopy from the ground (grass and soil). We will discuss the implications and applications for fusing multispectral and SfM derived information to characterise the variation of structural and biochemical plant traits across the heterogeneous scene. We will also present a technical workflow for coupling drone derived information (e.g. canopy height model and LAI estimates) with field measurements for the spatial parameterisation of a 3D radiative transfer model (Discrete Anisotropic Radiative Transfer Model, DART) in terms of LAI and voxel height and distribution for simulations in forward mode. The potential and limitations of this approach for use e.g. in a spectral sensitivity analysis or inversion framework, and when considering its application to more complex ecosystems will be highlighted. Finally, the steps in the workflow to integrate observations from the ground, through drone and airborne data, to the satellite observations will be discussed.
As an outlook, the presented work may have implications for scientists seeking to use more coarsely resolved (10-30 m) optical observations to derive information on smaller scale processes (e.g spatial ecology).

Area A - Board 56